GitHub Copilot for Azure is a GitHub Copilot extension that enables developers to use natural language to:

• Learn about Azure features.
• Deploy Azure resources.
• Get information about Azure resources.
• Diagnose and troubleshoot problems with Azure resources.

You must have access to an Azure subscription and be subscribed to GitHub Copilot. Get started using the extension.

GitHub Copilot is designed to help developers, including developers new to Azure, to be more productive as quickly as possible. For experienced Azure users, GitHub Copilot for Azure replaces the need to:

• memorize or look up Azure CLI commands and arguments.
• create complex deployment scripts by hand.
• sign in and browse through the Azure portal.

GitHub Copilot for Azure supplements the general knowledge of the latest foundational large language models (LLMs) with tool calling by using the Azure Model Context Protocol (MCP) Server. This server enables interaction with Azure services, systems, and Azure Resource Graph to carry out specific tasks on your behalf. Dozens of Azure services are already available, and more services and capabilities are being added regularly. See the full list of tools for details. Learn more about the capabilities of Azure MCP Server.

Azure Skills - including azure-prepare, azure-validate, and azure-deploy - are also included and install automatically with the extension, enabling structured multi-step workflows for common Azure tasks.

You can use GitHub Copilot for Azure in the following supported development environments:

GitHub Copilot for Azure currently enables four primary scenarios:

The documentation provides a quickstart and example prompts to help you start using GitHub Copilot for Azure as quickly as possible.

First, use Agent mode for the best experience when completing multi-step tasks. Use Ask mode (@azure) for quick questions about Azure services.

Second, include the word "Azure" in the prompt to help Copilot understand that it needs to call tools from the Azure MCP Server.

Third, using copilots can increase developer productivity by answering questions, executing tasks, and generating code. However, remember these vital rules:

• Review all AI-generated responses. Validate their correctness, applicability, potential outcomes (such as costs and security) before taking action based on those responses.
• Never save application secrets or credentials in source code.
• Never submit application secrets or credentials in questions or in code when you ask questions.

When you're working with any tool based on large language models use good prompt engineering techniques for the best results. The following tips come from the article Write effective prompts for Microsoft Copilot in Azure, which provides advice for prompt engineering in the context of Azure.

• Be clear and specific
• Set expectations
• Add context about your scenario
• Break down your requests
• Customize your code
• Use Azure terminology
• Use the feedback loop

When working in agent mode, you can create longer prompts, however it's important to constrain the copilot before allowing it to act on your behalf especially when working with your Azure account. Here's an approach to building a longer prompt that might help get the results you desire.

• Command - "Don't take any action until I authorize it." Prevent the copilot from taking action before you validate its understanding of the prompt.
• Describe - Express what you want to happen. Here, you describe the work like you would to a coworker in sufficient detail for your coworker to be successful.
• Ask - "Do you have any clarifying questions to ask me before you begin?" - Give the copilot an opportunity to identify unclear instructions.
• Iterate - Iterate with the copilot until it understands what you're asking it to do. The copilot might require several iterations before it has everything it needs to be successful.
• Request - "Create a step-by-step checklist plan that I can review before I authorize you to execute the plan." This not only forces the copilot to think ahead of its actions and explain its approach, it also follows these steps and provides a status.
• Review - At some point, you might trust the copilot and not closely review its work. However, it's always best to make sure you review the plan and clarify what you want.
• Authorize - "I reviewed the plan and you're authorized to begin."
• Validate - Spend time checking the work to ensure that it accomplishes what you intended.

GitHub Copilot for Azure uses agentic tools behind the scenes to perform all operations. When first released for Visual Studio Code, the GitHub Copilot for Azure team created proprietary tools for use, however the current direction is to migrate away from proprietary tools to tools supplied by the Azure MCP Server.

Here's a list of all the tools currently supported by GitHub Copilot for Azure. Also listed is the migration status, which applies solely to the Visual Studio Code version. The Visual Studio 2022 version uses Azure MCP Server tools by default.

GitHub Copilot for Azure uses agentic tools supplied by the Azure MCP Server behind the scenes to perform all operations. The tools appear in all tools mode and need to be manually enabled before use.

GitHub Copilot for Azure uses agentic tools supplied by the Azure MCP Server behind the scenes to perform all operations. In addition to Azure MCP Server tools, the following Visual Studio-specific tools are also available.

• Example prompts for learning about Azure and your application with GitHub Copilot for Azure
• Example prompts for designing and developing your application with GitHub Copilot for Azure
• Example prompts for deploying your application with GitHub Copilot for Azure
• Troubleshoot Azure applications with GitHub Copilot for Azure
• Get started with GitHub Copilot for Azure
• Quickstart: Deploy your application to Azure with agent mode in GitHub Copilot for Azure
• Generate Bicep files using GitHub Copilot for Azure